Cluster Admin at soket.ai (Bengaluru, India)

soket.ai· Bengaluru, India·15h ago

Role details

Work type

Onsite

Employment

Full-Time

Job description

Some

Mission-Driven Culture: Opportunity to work on frontier AI (math, code, reasoning) with an explicit ethical mandate—rare in the for-profit AI lab space.
High-Impact Work: Directly contribute to India’s sovereign AI infrastructure, with work presented to national leaders and global tech giants.
Research-Heavy Environment: 38% of staff in technical roles; emphasis on deep research, open-source contributions, and building from scratch.
Hybrid Work Model: Offices in Gurugram (HQ) and Bengaluru; hybrid policy with a mix of remote and on-site work.
Growth Trajectory: Small, fast-growing team (30% YoY headcount growth) with active hiring in AI Research, Engineering, and Infrastructure—early-stage equity-like impact.
Notable Perks: Collaboration with Google Cloud, access to high-end compute (NVIDIA GPUs, Slurm clusters), and a seat at the table in India’s AI policy discussions.

One-liner: Soket AI Labs is an Indian research lab building sovereign, ethical frontier AI—including large language models, speech recognition, and reasoning systems—for India and the Global South.
Entity Type: Private (Bootstrapped/Pre-Series; funding details not publicly disclosed)
Headquarters: Gurugram, Haryana, India
Founded: 2019
Founders: Abhishek Upperwal (CEO & Founder)

Primary Industry: Artificial Intelligence Research & Development
Target Customers: Researchers, developers, enterprises, and government bodies in India and the Global South; B2B and B2G focus.
Mission: To build AGI with a conscience—advancing frontier math, code, and reasoning in parallel with foundational AI for Indian and Global South languages, built on principles of efficiency, ethics, and equity.

Project EKΛ (EKA): India’s sovereign foundation model initiative under the IndiaAI Mission. Includes the EKΛ Code 24B model (a 24B parameter code model) and a Sparse MoE architecture with 120B+ parameters. Open-source and focused on math, code, and reasoning.
Dhrith: A next-generation Automatic Speech Recognition (ASR) model designed for multilingual, code-switched speech transcription. Understands emotion, rhythm, and linguistic nuance across Indian languages.
Pragna-1B: A 1.25 billion parameter open-source multilingual foundational model built in collaboration with Google Cloud. Supports Hindi, Gujarati, Bangla, and English for NLP, translation, and conversational AI.
SARTHI: A Hindi-native agronomic reasoning model for agriculture, demonstrated to India’s IT Minister.
Realtime Speech API (Voice AI API): A real-time voice AI API for transcription and custom voice applications.
EKΛ Tokenizer: An open-source tokenizer built for Indian and Global South languages.

Valuation/Market Cap: Not publicly disclosed.
Key Metric: Total headcount of ~11-30 employees (sources vary; LinkedIn shows 11, BuiltIn shows 30); LinkedIn followers: 6,601 (+161.2% YoY).
Notable Investors/Partners: Google Cloud (collaboration on Pragna-1B), IIT Roorkee (research collaboration), NASSCOM, Yotta Data Services, Qualcomm, Adani Defence and Aerospace (showcased at shared events).
Growth Signals: Presented EKΛ Code 24B to India’s Prime Minister Narendra Modi; shared stage with Google, OpenAI, Meta, Anthropic, and Microsoft at the New Delhi Frontier AI Impact Commitments; strong LinkedIn follower growth (161% YoY); active hiring for 8+ roles across AI Research, Engineering, and Infrastructure.

Sovereign AI Focus: Building India’s own frontier models, data pipelines, and infrastructure—positioned as a national strategic asset under the IndiaAI Mission.
Open-Source Ethos: All models (Pragna-1B, EKΛ) are fully open-source, built in public, fostering community trust and adoption.
Multilingual & Global South Expertise: Deep focus on 22 Indian languages and 20 Global South languages, addressing a massive underserved market.
Ethical AGI Mission: Explicit commitment to humane, transparent, and energy-efficient AI development—differentiates from purely profit-driven labs.

Scaling Project EKΛ: Advancing toward a 120B+ parameter Sparse MoE foundation model as India’s sovereign AI brain.
Deepening Research Partnerships: Collaborating with IIT Roorkee and other academic institutions on datasets, benchmarking, and training.
Commercializing Speech & Voice AI: Expanding Dhrith ASR and the Realtime Speech API for enterprise and government use cases.
Government & Policy Engagement: Positioning as a key player in India’s AI sovereignty narrative, with high-level government visibility.

Sources